Non-Invasive Diagnosis of Deep Vein Thrombosis From Ultrasound With Machine Learning

Background: Deep Vein Thrombosis (DVT) is a blood clot most found in the leg, which can lead to fatal pulmonary embolism (PE). The diagnostic standard for DVT is a compression ultrasound examination of the legs. Many patients with possible symptoms are not found to have a DVT, resulting in long waiting times for patients and a large clinical burden for specialists. Machine learning has shown great potential for the automatic interpretation of images but has not been evaluated at the front line of care yet. We collect images in a pre-clinical study and investigate a machine-learning software for the interpretation of compression ultrasound images to aid non-specialists in detecting DVT. Methods: Selected anatomical areas are commonly scanned to examine for DVT (common femoral vein, superficial femoral vein, popliteal vein). A machine learning model automatically identifies the veins and arteries and predicts which of the areas, if any, is present in the ultrasound images. Compressions are algorithmically instructed, and compressibility is automatically evaluated. We tested on ultrasound videos scanned by professionals, where each frame has been manually labelled with the correct anatomical location and vessel compression status to quantify method performance. A cost-effectiveness analysis for clinical pathway integration is also provided. Findings: Evaluation was performed on a sample size of 51 prospectively enrolled patients from an NHS DVT diagnostic clinic. 32 DVT-positive patients and 19 DVT-negative patients were included. Algorithmic DVT diagnosis results in a sensitivity of 93.8% and a specificity of 84.2%, a positive predictive value of 90.9%, and a negative predictive value of 88.9% compared to clinical gold standard. Integrating our approach into a clinical DVT pathway is estimated to be cost effective at up to $150 per software examination, assuming a willingness to pay $26 000/QALY. Interpretation: Machine learning methods have good diagnostic accuracy and have the potential to be cost-effective for identifying DVT in patients. Future work will evaluate if front-line-of-care workers can accurately diagnose DVT with our approach. Funding: ThinkSono Ldt. (ThinkSono Ldt funded the development of the method. Data was collected at Oxford Haemophilia & Thrombosis Centre independently. ThinkSono Ldt provided ultrasound data acquisition devices to Oxford Haemophilia & Thrombosis Centre for this study.) Declaration of Interests: B.K., M.H. and N.C. are scientific advisors for ThinkSono Ldt B.K. is also advisor for Ultromics Ldt And Cydar medical Ldt J.O. acts as a consultant for ThinkSono Ldt. A.M. was an employee of ThinkSono Ldt until September 2020. F.AN., S.M. are employees of ThinkSono Ldt, M.D.S. and A.C.R. acted as contractor for ThinkSono Ldt. All authors had full access to all data during this study and accept responsibility to submit for publication. B.K., A.M., F.AN., S.M. are joint inventors on a patent held by ThinkSono Ldt. The views expressed are those of the author(s) and not necessarily those of ThinkSono, the NHS, the NIHR or the Department of Health. Ethics Approval Statement: The University of Oxford, UK, approved the study (Ethics: 18/SC/0220, IRAS 234007). All participants provided written informed consent.